Closed samgelman closed 2 years ago
This is a simple example showing how to run GPU jobs without a docker container. It uses Anaconda to install the required CUDA libraries and set up TensorFlow. This example uses a server whitelist, but in the future it could be updated to use GPU/CUDA properties. The script inside of my_tensorflow_program.py is from StackOverflow, and there's a reference URL in the comments of that file.
I recommend that we modify the scope of this example to be specifically for the GZK servers. There may not be high demand, but the example already runs so I'd like to document the process. That will simplify the submit file and executable substantially. The submit file can filter on properties of the GZK servers (is the OS sufficient?).
We could later make a new TensorFlow conda example that copies #8. It would only need modifications to the conda packages that are installed and the demo code that is run.
@samgelman I'm closing this old pull request because we now have general examples of creating conda environments with GPU support. I renamed the pull request so that we can find and reference these instructions later when others want to run on GZK servers and need the special setup tricks.
Initial commit